<But we were talking about a rather closed setting here, i.e., the academic/research institutions who are members of I2.>
I know, but I'm relating that to commercial applications. I know the rules change on dedicated networks (Those are the guys who say "I ain't waiting on no stinkin' network.") My point is in the last paragraph, below, but first some stream of thought.
I tend to relate everything back to the consumer level, and the experience I will have in getting on. (And my little 28.8 just keeps walking along...).
But even within the institutional setting, I think the same constraint applies--albeit to a lesser extent. If I understand developments in the switch/router market, things are not keeping up there with the bandwidth capabilities of the backbone. Constraints will be peculiar to each institution's infrastructure, and within each institution to each individual's hookup. Some departments will be directly connected to the 10/100 type lan. Others will be through hdsl lines. Etc. All will come together, presumably, at a junction box (switch? dslam? Probably a router.) with capacity lower than the speed of the trunk it's feeding into, and with a limited passthrough capability.
The situation's not much different from a PC, where the hard disk and busses and CPU each work at different speeds and with different capacities. One of the new Pentium III's isn't going to make last year's PC any faster.
And I now remember that you (?) used the on-ramp metaphor. The freeway might be moving along lickety-split, but if the on ramps are clogged up nobody will get to work any earlier.
To take it back to the consumer level again: I tried ISDN last year, after Bell Atlantic introduced it. That should have been a 4X speed up. NOT! Problems were (1) I was getting a busy signal in getting on--speed nil; (2) the server response appeared slower than with the 28.8 modem. This presumably related to the number of busy lines trying to use a limited number of servers. (3) Doing a tracert, you could see multiple 200 ms delays as requests went out through the routing process. And most of the time it was taking well over 10 steps to reach a destination--and then the same number back.
For instance, I have a site up that is being hosted in California (cheap per month rate, small space, reliable connection). It takes 15 steps to reach the site: 4 just to get out of Bell Atlantic's farm; 4 to nynap (on Iconnet), 3 to mae west, 2 to the hosting company and 2 within the hosting company to the site. Each step takes 130 to 220 ms, off peak--I just checked it. While Bell Atlantic is probably less efficient (takes more steps) than most in getting onto the backbone, I assume all commercial carriers are within this order of magnitude. To get to VoIP or videoconferencing the number of steps has to be reduced, as well as the time through each router/switch.
Anyway, all above offered FWIW.
Best,
JS |
